Skill balance and entrepreneurship evidence from online career histories.
We conduct a matched case-control analysis of transitions into entrepreneurship using the curriculum vitae (CVs) of individuals with ties to the venture capital and private equity (VCPE) industries. Our goal is to undertake another examination of the proposal that entrepreneurs are jacks-of-all-trades (Lazear, 2004), whose diverse skills enable them to carry out the varied tasks involved in the creation of a successful business (in contrast, wage workers typically do more specialized work requiring a narrower set of skills). The degree to which an individual's skill set is balanced is not often observable. However, Lazear (2005) conjectures that individuals can invest in the accumulation of diverse skills through education or on-the-job training. Individuals who intend to pursue an entrepreneurial career choose to learn a variety of skills, either by studying a varied curriculum in college or by accepting a variety of positions in work. Similarly, individuals who just happened to have varied educational or employment experiences will find themselves more fit for entrepreneurship, and hence are more likely to become an entrepreneur should any opportunity to do so arise.
Lazear's (2004) theory of entrepreneurship is well known, sufficiently so that it might now be considered one of the canonical economic models of entrepreneurship. As a consequence, there has been a growing literature focused on testing its key implications. Consider first the evidence on entrepreneurial entry. Lazear (2005) himself provided supporting evidence from the work histories and university transcripts of Stanford MBA alumni. Alumni who are entrepreneurs had studied a more diversified MBA curriculum than wage and salary workers, and they had a greater variety of roles in the labor market prior to becoming an entrepreneur. Among the earliest of empirical tests of the theory, Wagner (2003,2006) shows that diversity of roles in the labor market is positively associated with self-employment in a large random sample of the German labor force. (1) However, subsequent research on large German data sets has produced more nuanced results. For example, Lechmann and Schnabel (2014) find that the likelihood of business creation increased with the prior number of changes in profession only for entry into selfemployment; they found no relationship for the creation of business that employed other workers. Moreover, they found no relationship between the number of different kinds of professional training and the probability of being self-employed.
Silva (2007) produces mixed evidence from Italy. He shows that graduates of Bocconi University who followed a balanced curriculum were more likely to become entrepreneurs. However, he also reports that the number of prior roles held by individuals in the Longitudinal Survey of Italian Families is positively associated with the likelihood of becoming an entrepreneur only in the cross section; no pattern was found with panel estimates. Silva interprets the cross-sectional results as the consequence of selection on unobservables, while the panel estimates lead him to reject the notion that job-hopping reflects investment in skill accumulation. Astebro, Chen, and Thompson (2011) obtain similar results with a large Korean panel: they find that variety is positively associated with entry into entrepreneurship in the cross section but not in panel estimates.
The jack-of-all-trades theory also has implications for the relationship between work histories and performance, but the evidence on this score is similarly mixed. Studies of nascent entrepreneurs have generally shown a positive relationship between experience variety and earnings among entrepreneurs in the United States (Hartog, van Praag, & van der Sluis, 2010; Stuetzer, Goethner, & Cantner, 2012; Stuetzer, Obschonka, & Schmitt-Rodermund, 2013), in Korea (Astebro et al., 2011), in Sweden (Toft-Kehler, Wennberg, & Kim, 2014) and in Germany (Bublitz & Noseleit, 2014). However, Bublitz and Noseleit find in German data that skill balance is associated with higher earnings among business owners only when the average skill level is high. Similarly, Roberts, Negro, and Swaminathan (2013) find no support in a study of restaurant owners for the prediction that serial entrepreneurs, who presumably have prior experience with the varied tasks of creating and running a business, obtained better customer reviews than first-time owners; to the contrary, owners who were also head chefs performed worse if they had previous ownership experience.
Astebro and Thompson (2011), analyzing a simple formal model of the theory, show that on average varied experience predicts higher earnings among entrepreneurs but not among employees. However, their empirical analysis of Canadian inventors shows a negative effect of variety on the earnings of both entrepreneurs and wage workers. They conclude that job- and industry-hopping reflects an individual taste for variety rather than an investment in skill accumulation. Hessels, Brixy, Naude, and Gries (2014) study nascent entrepreneurs in the Dutch and German subsamples of the Global Entrepreneurship Monitor. They find some evidence that respondents with more varied job histories were more likely to experience start-up success. However, respondents who considered themselves to be generalists were no more likely to succeed than others. Hessels et al. warn that a more varied job history need not correspond with having a more balanced skill set. (2)
Part of the reason for the mixed results may well be the diverse contextual settings in which empirical work has been conducted. Lazear (2005), and Astebro and Thompson (2011), for example, use retrospective surveys on rather specialized samples that may have limited external validity. In more representative samples, it is possible that data constraints limit internal validity. Notably, cross-sectional studies struggle to distinguish between Lazear's hypotheses that individuals invest in balanced skills prior to entry in entrepreneurship, and selection mechanisms whereby individuals more likely to become entrepreneurs happen to accumulate a more varied work history. Fixed-effects panel estimates, which eliminate the confounding effect of time-invariant individual characteristics that predict both entrepreneurship and job-hopping, resolve one selection mechanism. Their failure to identify direct effects of variety on entrepreneurship might be interpreted to favor selection mechanisms over Lazear's investment story. However, the panels studied tend to be fairly short, so there is frequently little within-individual variation in available measures of job variety. Moreover, panel studies are often constrained by coarsegrained classifications of jobs and industries, such that many substantive job changes within employers are not observable, while observed changes of employer or industry are not always substantive changes in jobs. This course-grained classification also suppresses heterogeneity in employers that adds to the noise found in panel estimates.
In this paper, we use fine-grained data from CVs uploaded to a professional online networking site, and construct a sample that allows us to assess whether job history influences the rate of entrepreneurship among individuals holding the same or a similar job, at the same firm, and at the same time. Because preprocessing CVs is very labor intensive, we construct our data using choice-based sampling. We first selected individuals who had identified themselves as being currently involved in the VCPE industry, and had at some time in their career founded a business. We then identified the last paid job they held before becoming a founder and searched for other individuals, to be used as controls, reporting a similar job title with the same employer around the same date as the founder. Our sample consists of 409 cases and 1,463 matched controls although, as will be discussed in Methods section, a fraction of the controls also became entrepreneurs.
Our data have some attractive features. First, they allow us to track changes in an individual's job responsibilities within a firm, and to verify whether changes of employers really entail a change in responsibilities; indeed, in many cases, lines in the CV are supplemented with extensive explanations of job responsibilities and accomplishments. Second, the data are nonanonymous, so it is frequently possible to identify a sample of coworkers, as well as to examine details about the employer using external sources of information. Third, the structure of CVs allows us to identify work activities conducted simultaneously, such as paid work among individuals creating a business, or service on boards of directors. Finally, we have access to precise information about geographic relocations, colleges attended, and college major. Thus, our data provide count measures of job variety comparable to the best cross-sectional specialized samples, an ability to dynamically trace career histories comparable to the best panel data sets, and an ability to control for unobserved job and employer characteristics comparable to the most detailed matched employer-employee databases.
Of course, all data sets involve compromises. First, we also have only limited demographic data (e.g., age must be inferred from dates of educational milestones, and gender from names and, perhaps, images). Second, there are no data on income so, while we can look at the impact of job histories on entry into entrepreneurship, we cannot look at their consequences for performance. Finally, our data are self-reported and may include inaccuracies. On the one hand, there may be a certain amount of embellishment, as network members attempt to secure better jobs and other professional opportunities; on the other hand, there may be omissions if members choose only to report those parts of their career they feel will best help achieve their current goals. However, Ge, Huang, and Png (2014) have used a survey instrument to assess the quality of CVs culled from an online network, and found that they were generally high quality.
Construction of Founder Set. We constructed our data set from one of the largest professional social network websites in the world. At the end of 2014, the website reported having more than 300 million members across more than 200 countries, with 100 million users in the United States alone. Users are able to construct online CVs, to interact with other professionals, and to search individuals based on individual qualifications or company associations. Users generally provide a summary of their background, work history, education history, and other information, such as organization associations or certificates. Users can also specify their current industry and location (e.g., Management Consulting in Great Los Angeles Area) to facilitate networking with other professionals.
We constructed our sample between July 2012 and November 2013. We first identified entrepreneurs by searching for individuals with "founder" in their reported career history. However, because of the huge user base, the search algorithm limits the number of profiles returned per search and prioritizes them in terms of "closeness" to the user who conducts the search. (3) This can be overcome by specifying zip-code searches. Hence, we built a pool of founders located in each of 250 zip codes in major cities covering 40 U.S. states and Canada. (4) To cover as large area as we could, we searched for founders within 100 miles of a given zip code, (5) which yielded 25,552 founders. We then focused on founders who currently identify themselves as working in the VCPE industry, resulting in 629 founders.
We did not require that these founders had started a business in the VCPE industry. However, we did attempt to identify founders of a "serious" business, by which we mean commercial ventures rather than those that appear to be self-employment; student ventures; and businesses inherited or bought from earlier owners, company branches, or regional chapters of a club. To identify the types of founders in which we are interested, we reviewed the founding information provided in the career histories and searched on the Internet for information about the listed venture, including the description of the venture, company website, and the founding year. We excluded for example, ventures that are founded and active only during the period when the focal founder is in school, ventures that offer only a blog-like website without clear business activity, and ventures that are branch offices of an existing firm.
We focused on the first founding event in a founder's career to avoid potential confounding effects from past founding experience. In addition, since we are interested in how the characteristics of pre-founding experience affect entrepreneurship, and we need to identify controls during that pre-founding period, we excluded founders without any pre-founding employment experience from our primary analysis. Individuals who do not report a complete career history may also affect our evaluation of accumulated skills. Hence, we excluded profiles that have a gap greater than 3 years. We ended up with 388 founders for whom we attempted to identify matched controls.
Construction of Matched Sample. We used each founder's employment experience immediately prior to his or her first venture to construct the matched control pool. A matched control for a founder is an individual who worked in the same company around the same time and was at a similar job level as the focal founder. For instance, if a founder worked in Accenture from 1997 to 2001 as a consultant before he started a business, we consider as a match any individuals who worked in Accenture during that four-year window with titles such as "Analyst," "Consultant," or "Senior Consultant," all of which are nonmanagement positions. We excluded founders without any matched controls, reducing the sample size of founders to 324. There are in our final sample 1,548 matched controls, an average of 4.78 controls per founder with a range from 1 to 31.
Data and Measurement
Dependent Variable. We analyzed a person's experience accumulated up to the year before the first business venture of the associated founder. We measured an individual's transition to entrepreneurship by coding an indicator variable as one if the person is a founder of the company and as zero otherwise. Since some individuals in the control group transitioned to entrepreneurship shortly after the date used for matching, we created a variable that is coded 1 if an individual founds a company within 3 years of the reference year, and 0 otherwise. Under this 3-year window, 85 individuals from the control group are also entrepreneurs. As a result, the analysis sample consists of 409 entrepreneurs and 1,463 nonentrepreneurs.
Number of Functional Experiences. We followed studies in jack-of-all-trades and new venture performance (Li & Zhang, 2007; McGee, Dowling, & Megginson, 1995; Stuetzer, Obschonka, Davidsson, et al., 2013) and categorized the reported job experience into six functional areas: (1) accounting and finance, (2) business administration, (3) marketing and sales, (4) R&D and engineering, (5) personnel, and (6) production (see the Appendix). We then counted the number of different functional areas an individual has experienced prior to the reference year. The jack-of-all-trades theory predicts that individuals with experience in more functional areas are more likely to become entrepreneurs.
Number of Prior Employers. Using data from the Korean Labor and Income Panel Study, Astebro et al. (2011) found that the number of prior employers was positively related to the likelihood of entering self-employment. They argue that Lazear's (2004) theory offers one possible explanation for this finding because job hopping may be related to underlying skill balance. Therefore, we counted the number of distinct prior employers prior to the reference year.
Controls. We reviewed profile pictures, names, bibliographies of subjects in the sample, and information on the Internet to determine sex. The variable female is coded as 1 if the person is female and 0 otherwise. Data on age is not available to us, so we inferred a subject's age from his/her reported profile. We assumed that an individual is 22-years-old at the time of obtaining a bachelor's degree. If a person reports the years of his/her degree, we used the reported year to infer the age. Some people report their degree and school names only or do not provide any education history at all. We reviewed their profiles and determined the appropriate age when possible and excluded them from the sample when not. Since previous studies often found a nonlinear relationship between age and entrepreneurship, (e.g., Rider, Thompson, Kacperczyk, & Tag, 2013), we added age squared to the model.
We created four education variables. We coded the variable Bachelor as 1 if a person had a bachelor's degree. Similarly, the variable MBA takes a value of 1 if a person had an MBA degree before founding their first company. We also investigated the reported education history and identified business-related subjects, such as Business, Management, Finance, Accounting, Management Information Systems, and so on. We coded the variable Business Education as 1 if a person had any business-related education, regardless of whether it led to a degree. Finally, the variable Entrepreneurship Focus takes a value of 1 if a person reported an entrepreneurship concentration in his/her education history.
To account for individual quality and the effect of career progression, we created three binary variables related to job rank, Board Experience, Senior Management Experience, and Rank Loss. Experience as a board member may identify individuals with high ability or more extensive networks, both of which may increase the likelihood of entrepreneurship. The variable Board Experience was coded as 1 if a person in our sample reported appointment to a company board. Blanchflower and Oswald (1998) suggested that low aspiration could lead to self-employment. Hence, we created the variable Rank Loss to account for the effect of a decline in rank. We classified the reported job title into four job ranks: nonmanagement, submanagement, management, and top management, which are coded from 1 to 4, respectively. The variable Rank Loss is constructed by subtracting the job rank an individual held immediately prior to the reference year from the highest job rank an individual had ever attained. If an individual did not experience demotion, this variable is 0. On the other hand, if an individual was demoted in terms of job ranks, for example, from management level to submanagement level, this variable is positive. Finally, we coded for two types of experience, Accounting and Finance Experience and Business Administration Experience, which are particularly prominent in our sample. Finally, the variable Senior Management Experience was coded as 1 if a person in our sample reported holding a position at the management or top management level. All three variables are binary and take a value of 1 if an individual had the indicated experience before the reference year.
Industry of Venture. We assigned a two-digit NAICS code to all ventures in our sample according to the description of the venture, its company website, or business registration information available on the Internet. We used the 2012 industry sector definition provided by the North American Industry Classification System. (6) Figure 1 shows the industry distribution of the 409 ventures in this study. The top three sectors are Information (N = 128); Finance and Insurance (N = 103); and Professional, Scientific, and Technical Services (N = 102). The three sectors account for 80% of the ventures in the sample. Among the 103 ventures in the Finance and Insurance sector, 66 firms engage in VCPE activities, accounting for 16% of the new ventures.
[FIGURE 1 OMITTED]
Dropped Case Analysis
We conducted tests on the 305 founders from our initial sample draw to see if there was any difference between the dropped cases and our final sample. We first tested the gender composition. There were 279 men and 26 women in the dropped cases. The Chisquared test of independence indicates that there is no statistical difference in the gender compositions between the two groups (p = 0.26). We then compared the age distribution between the two. There were 44 profiles for which we cannot infer the individual's age. We allocated the remainder to one of five age groups to avoid issues of small cell size (Agresti, 2013). The p-value from the Chi-squared test was 0.000, indicating a statistical difference in the age distribution. The most notable difference is that the dropped cases have 53 individuals younger than 20 years old, compared with only two in the final sample. This result is unsurprising, of course, given our requirement that entrepreneurs have pre-founding experience on which we can match. We also compared the industry distribution between the two groups. We classified observations into six industry groups, again to avoid small counts in some sectors. The Chi-squared test indicates a marginally statistical difference in the industry composition (p = 0.09). The final sample has a greater representation in the information and finance sectors, while the dropped cases have a more diverse industry distribution. Note that these are all founders who identify themselves in the VCPE industry at the time of data collection. Since incomplete career histories account for about two-thirds of the dropped cases), it is likely that users in the information and the finance sectors are more familiar with online self-presentation.
Table 1 provides the summary statistics of the complete sample. Twenty percent of the samples are women, but only ten percent of the entrepreneurs are female. Mean age in the reference year is 31, and there is no difference between the founder and control samples: our matching on job title and employer also yields a similar age profile. This is a young sample, likely because younger people are more likely to use social media. As a result, the mean number of prior employers, 2.94, is rather small. The entrepreneur group has slightly more job switching experience than the matched sample group. In addition, the population has on average 1.62 functional skills before the first founding event. About 80% of the sample has a bachelor's degree. While 47% of the population received business-related education, only 25% had obtained an MBA before the reference year. In terms of industry-related characteristics, 39% of the population had engaged in tasks related to accounting and finance and 41% had experience in business administration.
Table 2 shows the correlation matrix for the sample as a whole. The correlations between the entrepreneurship indicator and other variables suggest no notable monotonic relationship in the raw data, except that women are less likely to become entrepreneurs. There are, however, several notable correlations that will prove relevant to our subsequent regression analysis. First, the number of prior employers and number of functional experiences are highly correlated with each other (r = 0.48), and both are correlated with age (r = 0.48 and 0.41). We will therefore flexibly control for age in all regressions, and we will (following Astebro et al., 2011) attempt to discriminate between the separate contributions of functional and employer experiences. Second, indicator variables for specific functional experiences are positively correlated with the count of functional experiences; this is not surprising, but it suggests that we should examine whether it is the variety of experiences or the fact that an agent has some experience in particular that matters for entrepreneurship.
We have a case-cohort sample in which the incidence of entrepreneurship is much greater than in the population as a whole. The data further consist of J groups with sizes [k.sub.j], j = 1,2, In each group j, at least one observation (the focal observation around which the group was constructed) is of an entrepreneur; several of the remaining [k.sub.j-1] members of the group may also have been identified as entrepreneurs when using the 3-year window discussed in Methods section.
A natural choice of estimator in this context is the conditional logit (e.g., Chamberlain 1980),
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where [y.sub.ij]-- 1 if individual i in group j became an entrepreneur, and [y.sub.ij] =0 otherwise, and [x.sub.ij] is a vector of individual characteristics including career histories. Uniquely among common estimators, the logit yields consistent estimates of the vector [beta] despite the choice-based sampling (e.g., Prentice & Pyke, 1979).
Table 3 reports two sets of results. Panel A reports the results of ordinary logit regressions, the sort of analysis that would be carried out in the absence of a matched casecontrol sample. Panel B reports the results of the conditional logit estimation indicated by our sampling scheme. In each panel, we include the numbers of previous employers and functional experiences separately and jointly. (7) Consider first the results from Panel A. Columns (1) and (2) show that, when our key regressors are included separately, each has a significant and positive effect on the likelihood of entrepreneurship. The magnitude of the two estimated effects is similar, given that the sample variance of the number of functional experiences is somewhat larger than that of the number of employers. Column (3) reports the results when both variables are included in the logit regression. The coefficient on the number of functional experiences is greatly reduced and becomes insignificant, while the coefficient on the number of prior employers is unchanged from column (2). The logit regressions in Panel A therefore offer evidence that it is job hopping rather than balanced skills that predict entrepreneurship.
The results in Panel B tell a rather different story. In the first two columns, as in Panel A, the key regressors each have a sizeable and significant positive effect on the likelihood of entrepreneurship. Note also that the coefficient on prior employers is rather smaller than was the case in Panel A while, if anything, the coefficient on functional experiences has increased. When both regressors are included, in column (3), we find that functional experience matters more than prior employers (recall also that the former has a smaller variance than the latter). The natural inference is that unobserved firm effects lie behind the contrast in results between the two panels. That is, firms with conditions likely to induce employees to become entrepreneurs are more likely to hire workers who have a greater number of prior employers, but they are less likely to hire workers with varied functional experience.
Given our sample, the causes of these firm effects must remain unexplained. However, the fact that they are evident in our results highlights the value of the case-control method to account for them. In the remainder of this section, we build on the conditional logit results of Panel B and assess the extent to which the effects of skill and employer variety on entrepreneurship and can be explained by individual attributes.
In both sets of regressions, women are shown to be much less likely than men to become entrepreneurs in all models. The odds ratio implied by Table 3 suggests that ceteris paribus women in our sample are only 35% as likely as men to become entrepreneurs. We do not identify a clear relationship between age and entrepreneurship, which appears inconsistent with the previously documented inverse u-shaped relationship (e.g., Levesque & Minniti, 2006; Rider et al., 2013). This is possibly due to the relative youth of our sample.
Table 4 adds to our conditional logit regressions various measures of educational background, while continuing to include age and gender as controls. Having a bachelor's degree does not lead to entrepreneurial entry. Perhaps surprisingly, reporting a business-related education in personal profiles also has no significant relationship with entrepreneurship. On the other hand, and of course unsurprisingly, people who had an entrepreneurship focus in their studies, or had obtained an MBA, are more likely to become an entrepreneur.
Given how few of our educational measures appear to matter for entrepreneurship it is perhaps not surprising that their inclusion does not markedly alter the results already reported in Table 3 for our key variables. We continue to find that both the number of prior employers and the number of functional experiences have a positive effect on the likelihood of entrepreneurial entry, although there is some loss in precision in the estimated effect for functional experience.
In Table 5, we add a number of experience indicators. The regressions show that having experience in business administration, senior management experience, finance, and as a board member are each positively associated with entry into entrepreneurship. We also find a strong link between a loss of rank (demotion) and transition into entrepreneurship. The coefficient of Rank Loss is positive and statistically significant in all three columns, indicating that those who experienced some kinds of "demotion" in their career history are more likely to start their own business. Most important, given the focus of our study, is that the inclusion of specific types of experience that turn out to matter for entrepreneurship eliminates the positive effects of both employer and functional experience counts: in all three columns, the estimated coefficients are either much reduced in magnitude or the "wrong" sign, and all are statistically insignificant. We therefore conclude that what matters for entrepreneurship is not simply the range of experiences an individual has had, but the possession of certain specific types of experience. (8)
Lazear's (2005) jack-of-all-trades theory of entrepreneurship suggests that individuals with balanced skills are more likely to become entrepreneurs. Because founders need to work on diverse tasks by themselves, they must be good at a variety of tasks to be successful. Cross-sectional studies almost invariably confirm this theory, while panel studies have found, at best, ambiguous evidence. In this article, we examine the theory using a matched case-control sample constructed from online CVs, a study design that may account for unobserved heterogeneity of various forms, as well as exploit the richer information online career histories offer over traditional data sources. Our results demonstrate, first, that accounting for unobserved heterogeneity among employers can have important consequences for regression estimates. As a result, it is likely that cross-sectional estimates, and panel estimates with crude job classifications, are unlikely to be able to identify the effect of skill variety. Second, we find that while both employer and job variety are positively associated with entrepreneurship, the effect seems to be driven by the acquisition of certain specific types of experience.
While professional network data provide us the opportunity to design a matched case-control study using high-quality career history variables, these data have their own limitations. First, we are not able to obtain socioeconomic information, such as income, marital status, and family backgrounds, from the reported profiles. The matched casecontrol approach may ameliorate the problem because cases and controls are likely to be more similar to each other in terms of job qualification and socioeconomic status. However, it is still possible that some unobserved personal characteristics known to be associated with entrepreneurship, such as whether an individual's parents were themselves entrepreneurs, are also associated with the accumulation of balanced skills and of particular types of skill. The extent to which this undermines internal validity is a complex question. If there are unobserved factors that induce people to plan a career as an entrepreneur, and this in turn induces them to invest in the accumulation of a variety of skills, then this is exactly the sort of relationship between functional variety and entrepreneurship that Lazear's theory addresses. If, on the other hand, unobserved factors are associated both with functional variety and entrepreneurial aspirations because, say, individuals have a taste for variety (Astebro & Thompson, 2011), or have previously struggled to find a good job match (Astebro et al., 2011), then the results may contain a spurious component. The absence of income data in our sample makes it impossible for us to grapple with this question of causality, and is a limitation of using online CVs.
Construction of Functional Area Variables
We construct the functional area-related variables by looking for specific terms in people's reported job titles. Table Al illustrates a few terms that we use to categorize functional experience. We review the codified results from our algorithm and manually correct them if necessary. The complete categorization algorithm is available on request.
Table A1 Definition of Functional Areas Functional Areas Definition Accounting & Finance Job titles contain terms such as accountant, CPA, audit, or accounting, collateral, expense, budget, cash, payroll, billing, fund, default, debt, income, wealth, loan. stock, credit, financial, bank, capital, finance, banker, trader, broker, insurance. equity, portfolio, investment, securities, asset, or real estate Business Administration Job titles contain terms such as Management founder, owner, product manager, branch officer, consultant, consulting, strategy, corporate planning, or business analysis Marketing and Sales Job titles contain terms such as marketing, MKTG, MKT, sales, business develop-ment, account, or BD Personnel Job titles contain terms such as educator, compensation, HR, human resource, hir-ing, recruit, train, counselor, counseling, mentor, or learn Production Job titles contain terms such as operation, logistics, and supply chain R&D and Engineering Job titles contain terms such as engineer, research and development, R&D, pro-grammer, developer, research technician, architect, research, development, information technology, system admin, or network admin Table A2 Definition of Job Ranks Job Rank Definition Top Management Job titles contain c-level jobs, such as general manager, president, and principal Management Job titles contain director and vice- president Submanagement Job titles contain team leader and manager
Construction of Management Ranks
We categorize job titles into four job ranks: top management, management, submanagement, and nonmanagement. Work titles that do not belong to the three management categories are classified as a nonmanagement job. The classification rules are given in Table A2.
Agresti, A. (2013). Categorical data analysis (3rd ed.). Hoboken, NJ: Wiley.
Astebro, T., Chen, J., & Thompson, P. (2011). Stars and misfits: Self-employment and labor market frictions. Management Science, 57(11), 1999-2017. doi: 10.1287/mnsc. 1110.1400
Astebro, T. & Thompson, P. (2011). Entrepreneurs, jacks of all trades or hobos? Research Policy, 40(5), 637-649. doi: 10.1016/j.respol.2011.01.010
Backes-Gellner, U. & Moog, P. (2013). The disposition to become an entrepreneur and the jacks-ofali-trades in social and human capital. Journal of Behavioral and Experimental Economics, 47(C), 55-72.
Baum, RJ. & Locke, E.A. (2004). The relationship of entrepreneurial traits, skill, and motivation to subsequent venture growth. Journal of Applied Psychology, <59(4), 587-598.
Bublitz, E. & Noseleit, F. (2014). The skill balancing act: When does broad expertise pay off? Small Business Economics, 42(1), 17-32.
Blanchflower, D.G. & Oswald, A.J. (1998). What makes an entrepreneur? Journal of Labor Economics, 16(1), 26-60. doi:10.1086/209881
Chamberlain, G. (1980). Analysis of covariance with qualitative data. The Review of Economic Studies, 47(1), 225-238.
Chandler, G.N. & Hanks, S.H. (1994). Founder competence, the environment, and venture performance. Entrepreneurship Theory and Practice, 18, 77-89.
Ge, C., Huang, K.-W., & Png, I.P.L. (2014). Engineer/scientist careers: Patents, online profiles, and misclassification bias. Working paper. National University of Singapore. Available at http://ssrn.com/ abstract=2531477 or http://dx.doi.org/10.2139/ssrn.2531477, accessed 20 November 2015.
Hartog, J., van Praag, M., & van der Sluis, J. (2010). If you are so smart, why aren't you an entrepreneur? Returns to cognitive and social ability. Journal of Economics and Management Strategy, 19(4), 947-989.
Hessels, J., Brixy, U., Naude, W., & Gries, T. (2014). Skill variety, innovation and new business formation. IZA Discussion Paper No. 7889. Bonn: Institute for the Study of Labor (IZA).
Lazear, E.P. (2004). Balanced skills and entrepreneurship. The American Economic Review, 94(2), 208-211.
Lazear, E.P. (2005). Entrepreneurship. Journal of Labor Economics, 23(4), 649-680.
Lechmann, D.S.J. & Schnabel, C. (2014). Are the self-employed really jacks-of-all-trades? Testing the assumptions and implications of Lazear's theory of entrepreneurship with German data. Small Business Economics, 42(1), 59-76. doi: 10.1007/sl 1187-012-9464-6
Levesque, M. & Minniti, M. (2006). The effect of aging on entrepreneurial behavior. Journal of Business Venturing, 21, 177-194.
Li, H., & Zhang, Y. (2007). The role of managers' political networking and functional experience in new venture performance: Evidence from China"s transition economy. Strategic Management Journal, 28(8), 791-804. doi: 10.1002/smj.605
McGee, J.E., Dowling, M.J., & Megginson, W.L. (1995). Cooperative strategy and new venture performance: The role of business strategy and management experience. Strategic Management Journal, 16(1), 565-580. doi: 10.1002/smj.4250160706
Prentice, R.L., & Pyke, R. (1979). Logistic disease incidence models and case-control studies. Biometrika, 66(3), 403-411. doi: 10.1093/biomet/66.3.403
Rider, C.I., Thompson, P., Kacperczyk, A., & Tag, J. (2013). Experience and entrepreneurship. IFN Working Paper 970. Stockholm: Research Institute of Industrial Economics (IFN).
Roberts, P.W., Negro, G., & Swaminathan, A. (2013). Balancing the skill sets of founders: Implications for the quality of organizational outputs. Strategic Organization, 11(1), 35-55. doi: 10.1177/ 1476127012460944
Silva, O. (2007). The Jack-of-All-Trades entrepreneur: Innate talent or acquired skill? Economics Letters, 97(2), 118-123. doi: 10.1016/j.econlet.2007.02.027
Stuetzer, M., Goethner, M., & Cantner, U. (2012). Do balanced skills help nascent entrepreneurs to make progress in the venture creation process? Economics Letters, 117, 186-188.
Stuetzer, M., Obschonka, M., Davidsson, P., & Schmitt-Rodermund, E. (2013). Where do entrepreneurial skills come from? Applied Economics Letters, 20(12), 1183-1186.
Stuetzer, M., Obschonka, M., & Schmitt-Rodermund, E. (2013). Balanced skills among nascent entrepreneurs. Small Business Economics, 4/(1), 93-114.
Toft-Kehler, R., Wennberg, K., & Kim, P.H. (2014). Practice makes perfect: Entrepreneurial-experience curves and venture performance, Journal of Business Venturing, 29(4), 453^170. doi: 10.1016/ j.jbusvent.2013.07.001
Wagner, J. (2003). Testing Lazear's jack-of-all-trades view of entrepreneurship with German micro data. Applied Economics Letters, 70(11), 687-689. doi: 10.1080/1350485032000133273
Wagner, J. (2006). Are nascent entrepreneurs "Jacks-of-all-trades?" A test of Lazear's theory of entrepreneurship with German data. Applied Economics, 38(20), 2415-2419. doi: 10.1080/ 00036840500427783
(1.) Backes-Gellner and Moog (2013) have shown that German students who have a more balanced portfolio of social and human capital exhibit greater entrepreneurial aspirations.
(2.) However, there is direct evidence that varied job histories are associated with having (varied) skill sets associated with entrepreneurial success. Stuetzer, Obschonka, Davidsson, and Schmitt-Rodermund (2013) conduct a direct test of the relationship between job variety and ownership of the set of skills deemed helpful for entrepreneurial success. Using two well-established scales to measure self-reported entrepreneurial skills (Baum & Locke, 2004; Chandler & Hanks, 1994), they find that varied work experience is positively related to entrepreneurial skills while more traditional measures of human capital show inconsistent and statistically insignificant effects.
(3.) Individuals that have some association with the user who conducts the search are shown on the top of the result list.
(4.) The list of cities and states at the time of data collection is available from the authors.
(5.) The search function allows users to search for individuals by job title, company, zip codes, countries, industries, and so on. Searches typically yield several hundred records.
(7.) Additional controls will be introduced in later tables.
(8.) It is possible that the control variables "business administration experience" and "senior management experience," which may reflect jobs offering general experience, are picking up the kind of balanced skills that matter for entrepreneurship. However, when we estimate the models in Table 5 without either or both variables, the two focal variables in Table 5 remain statistically insignificant.
Li-Wei Chen is a doctoral candidate at Goizueta Business School, Emory University, 1300 Clifton Rd., Atlanta, GA 30322, USA.
Peter Thompson is a professor at Scheller College of Business, Georgia Institute of Technology, 800 West Peachtree Street NW, Atlanta, GA, 30308, USA.
Please send correspondence to: Li-Wei Chen, tel.: (404) 919-9813; e-mail: email@example.com and to Peter Thompson at firstname.lastname@example.org.
Caption: Figure 1: Industry Distribution of Ventures, N = 409
Table 1 Summary Statistics All samples Entrepreneurs Variable N = 1,872 N = 409 Mean Std. Dev. Mean Std. Dev. Entrepreneurship (0/1) 0.22 0.41 1.00 0.00 Female (0/1) 0.20 0.40 0.09 0.28 Age 31.38 7.47 30.83 7.15 No. of prior employers 2.94 2.22 3.26 2.69 No. of functional exp. 1.62 1.03 1.71 1.06 Bachelor (0/1) 0.79 0.41 0.81 0.39 Business education (0/1) 0.47 0.50 0.49 0.50 Entrep. focus (0/1) 0.01 0.10 0.03 0.16 MBA (0/1) 0.25 0.43 0.31 0.46 Rank loss 0.21 0.61 0.32 0.79 Board exp. (0/1) 0.04 0.19 0.07 0.26 Bus. admin, exp. (0/1) 0.41 0.49 0.47 0.50 Acct. and fin. exp. (0/1) 0.39 0.49 0.41 0.49 Senior management exp. (0/1) 0.49 0.50 0.59 0.49 Control sample Variable N = 1,463 Mean Std. Dev. Entrepreneurship (0/1) 0.00 0.00 Female (0/1) 0.23 0.42 Age 31.53 7.55 No. of prior employers 2.85 2.06 No. of functional exp. 1.60 1.02 Bachelor (0/1) 0.78 0.41 Business education (0/1) 0.47 0.50 Entrep. focus (0/1) 0.005 0.07 MBA (0/1) 0.24 0.43 Rank loss 0.19 0.55 Board exp. (0/1) 0.03 0.17 Bus. admin, exp. (0/1) 0.39 0.49 Acct. and fin. exp. (0/1) 0.38 0.49 Senior management exp. (0/1) 0.46 0.50 Table 2 Correlation Matrix (N = 1,872) Variable -1 (2) (3) (4) (1) Entrepreneurship (0/1) 1 (2) Female (0/1) -0.15 1 (3) Age -0.04 -0.09 1 (4) No. of prior employers 0.08 -0.05 0.48 l (5) No. of functional exp. 0.05 -0.01 0.28 0.41 (6) Bachelor (0/1) 0.03 -0.09 -0.001 -0.04 (7) Business education (0/1) 0.02 -0.04 0.06 0.07 (8) Entrep. focus (0/1) 0.09 0.01 -0.02 -0.002 (9) MBA (0/1) 0.07 -0.04 0.17 0.12 (10) Rank loss 0.09 -0.02 0.16 0.18 (11) Board exp. (0/1) 0.10 -0.01 0.24 0.40 (12) Bus. admin, exp. (0/1) 0.07 0.01 0.19 0.31 (13) Acct. and fin. exp. (0/1) 0.03 -0.05 0.02 0.08 (14) Senior management (0/1) 0.11 -0.12 0.47 0.35 Correlation Matrix (N = 1,872) Variable (5) (6) (7) (8) (9) (1) Entrepreneurship (0/1) (2) Female (0/1) (3) Age (4) No. of prior employers (5) No. of functional exp. 1 (6) Bachelor (0/1) -0.02 1 (7) Business education (0/1) 0.13 0.29 1 (8) Entrep. focus (0/1) 0.05 0.02 0.10 l (9) MBA (0/1) 0.14 0.19 0.62 0.12 1 (10) Rank loss 0.16 0.001 0.06 0.01 0.08 (11) Board exp. (0/1) 0.14 -0.001 0.03 -0.02 0.07 (12) Bus. admin, exp. (0/1) 0.55 -0.003 0.10 0.03 0.11 (13) Acct. and fin. exp. (0/1) 0.31 0.05 0.21 0.01 0.13 (14) Senior management (0/1) 0.21 -0.002 0.14 0.002 0.16 Correlation Matrix (N = 1,872) Variable (10) (ii) (12) (13) (14) (1) Entrepreneurship (0/1) (2) Female (0/1) (3) Age (4) No. of prior employers (5) No. of functional exp. (6) Bachelor (0/1) (7) Business education (0/1) (8) Entrep. focus (0/1) (9) MBA (0/1) (10) Rank loss 1 (11) Board exp. (0/1) 0.09 1 (12) Bus. admin, exp. (0/1) 0.16 0.16 1 (13) Acct. and fin. exp. (0/1) 0.05 0.03 -0.05 1 (14) Senior management (0/1) 0.19 0.17 0.16 0.07 1 Table 3 Models of the Likelihood of Becoming an Entrepreneur Within Three Years Panel A Logit Models Entrepreneurship (0/1) -1 (2) (3) No. of functional experiences 0.158 *** 0.081 (0.057) (0.061) No. of prior employers 0.130 *** 0.117 *** (0.028) (0.029) Female (0/1) -1.225 *** -1.223 *** -1.227 *** (0.188) (0.189) (0.189) Age -0.027 -0.029 -0.036 (0.052) (0.052) (0.052) Age squared 0.00004 -0.0001 -0.0001 (0.001) (0.001) (0.001) Panel B Conditional Logit Models Entrepreneurship (0/1) (1) (2) (3) No. of functional experiences 0.187 *** 0.147 ** (0.069) (0.073) No. of prior employers 0.087 *** 0.066 * (0.033) (0.035) Female (0/1) -1 271 *** -1.263 *** -1.27 1*** (0.204) (0.204) (0.204) Age -0.049 -0.043 -0.057 (0.062) (0.062) (0.062) Age Squared 0.0002 -0.00001 0.0002 (0.001) (0.001) (0.001) N= 1,872. Standard errors in parentheses. ***p<0.01, **p<0.05, *p<0.1. In panel A, the constant term is not reported. Table 4 Models of the Likelihood of Becoming an Entrepreneur Within Three Years (With Educational Controls) Conditional Logit Models Entrepreneurship (0/1) (1) (2) (3) No. of functional experiences 0.169 ** 0.122 * (0.070) (0.074) No. of prior employers 0.094 *** 0.076 ** (0.034) (0.036) Female (0/1) -1.281 *** -1.278 *** -1.283 *** (0.207) (0.207) (0.207) Age -0.077 -0.076 -0.086 (0.062) (0.063) (0.063) Age squared 0.001 0.0004 0.001 (0.001) (0.001) (0.001) Bachelor (0/1) 0.100 0.115 0.120 (0.165) (0.166) (0.165) Business education (0/1) -0.290 * -0.275 -0.289 (0.176) (0.177) (0.177) Entrep. focus (0/1) 1.320 ** 1.420 *** 1.369 ** (0.551) (0.549) (0.549) MBA (0/1) 0.546 *** 0.564 *** 0.552 *** (0.197) (0.197) (0.197) N= 1,872. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Table 5 Models of the Likelihood of Becoming an Entrepreneur Within Three Years (With Additional Controls) Conditional Logit Models Entrepreneurship (0/1) (1) (2) (3) No. of functional experiences -0.039 -0.054 (0.093) (0.0%) No. of prior employers 0.024 0.029 (0.038) (0.039) Female (0/1) -1.266 *** -1.265 *** -1.265 *** (0.210) (0.210) (0.210) Age -0.085 -0.091 -0.089 (0.066) (0.066) (0.066) Age squared 0.0004 0.0004 0.0004 (0.001) (0.001) (0.001) Bachelor (0/1) 0.137 0.147 0.145 (0.169) (0.170) (0.170) Business education (0/1) -0.371 ** -0.371 ** -0.371 ** (0.184) (0.184) (0.184) Entrep. focus (0/1) 1.409 ** 1.403 ** 1.426 ** (0.557) (0.554) (0.557) MBA (0/1) 0.551 *** 0.549 *** 0.553 *** (0.203) (0.203) (0.203) Rank loss 0.245 ** 0.247 ** 0.247 ** (0.101) (0.101) (0.101) Board exp. (0/1) 0.942 *** 0.886 *** 0.880 *** (0.306) (0.317) (0.318) Bus. admin, exp. (0/1) 0.423 ** 0.361 ** 0.413 ** (0.176) (0.151) (0.177) Acct. and fin. exp. (0/1) 0.334* 0.290 * 0.338 * (0.179) (0.157) (0.179) Senior management exp. (0/1) 0.510 ** 0.492 ** 0.495 ** (0.198) (0.199) (0.199) N= 1,872. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
|Printer friendly Cite/link Email Feedback|
|Author:||Chen, Li-Wei; Thompson, Peter|
|Publication:||Entrepreneurship: Theory and Practice|
|Date:||Mar 1, 2016|
|Previous Article:||Passing up the job: the role of gendered organizations and families in the entrepreneurial career process.|
|Next Article:||Do the self-employed more likely emerge from sequential or parallel work experience in business-related functions?|